Target Tracking With Movement Pattern Classification in Congestion

The patent segments the tracking process into multiple independent components: observation data processing, movement pattern classification, and track association. Each component handles specific aspects of target tracking independently, allowing accurate separation of multiple targets through classification of movement patterns (e.g., stationary, moving, crossing) while maintaining overall system manageability through modular architecture.

The patent introduces movement pattern classification as an intermediary layer between raw observation data and final track association. This intermediary component processes observation data through classification learning to generate movement pattern labels, which then serve as additional features for accurate track association, enabling separate tracking of multiple targets without directly increasing the complexity of the core tracking algorithm.

The patent performs preliminary action by pre-learning movement pattern classifications using past data before actual tracking occurs. The classification learning processing pre-computes the relationships between movement patterns and observation data features, creating a ready-to-use classification model that can quickly process new observation data during real-time tracking without requiring complex calculations at execution time.

The patent changes parameters by selecting and weighting specific movement pattern features based on their importance for target separation. Instead of uniformly processing all possible observation data features, the system identifies and emphasizes key parameters such as position changes, velocity vectors, and spatial relationships that are most discriminative for separating multiple targets, thereby reducing processing time while maintaining high separation accuracy.